Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
45
127
suite
stringclasses
4 values
task
stringclasses
38 values
demo
stringclasses
20 values
mode
stringclasses
1 value
slot
stringclasses
1 value
category
stringclasses
1 value
category_name
stringclasses
1 value
tier
stringclasses
1 value
solution
stringclasses
1 value
data_amount
stringclasses
1 value
verdict
stringclasses
1 value
right
bool
1 class
reason_available
bool
1 class
label_reason
stringclasses
13 values
circle_ok
bool
1 class
arrow_ok
bool
1 class
caption_ok
bool
1 class
consistency_ok
bool
1 class
sufficient_ok
bool
2 classes
severity
float32
1
1
caption
stringclasses
9 values
caption_level
stringclasses
2 values
n_referents
int32
1
7
n_destinations
int32
1
3
n_readings
int32
1
18
resolved_by
stringclasses
1 value
grounding_kind
stringclasses
1 value
circle_json
stringlengths
51
80
rendered_n_referents
int32
1
7
rendered_n_destinations
int32
1
4
rendered_n_readings
int32
1
18
render_verified
bool
1 class
arrow_json
stringlengths
102
109
note
stringclasses
103 values
source_frame
stringlengths
68
150
name_explicit
bool
1 class
image
dict
spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_0/C1#0
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_0
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
move this there
L3
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.3671875, "cy": 0.565234363079071, "r": 0.13371307373046876}
1
1
1
true
{"x0": 0.7453169900137082, "y0": 0.3653037490714137, "x1": 0.7599185992122239, "y1": 0.24441606908882374}
circle on flat_stove_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_0/frames/t0_rgb.png
false
{ "bytes": [ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 4, ...
spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_0/C1#1
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_0
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
put this there
L2
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.3671875, "cy": 0.565234363079071, "r": 0.13371307373046876}
1
1
1
true
{"x0": 0.7453169900137082, "y0": 0.3653037490714137, "x1": 0.7599185992122239, "y1": 0.24441606908882374}
circle on flat_stove_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_0/frames/t0_rgb.png
false
{ "bytes": [ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 4, ...
spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_1/C1#0
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_1
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
pick up this
L2
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.3671875, "cy": 0.565234363079071, "r": 0.13371307373046876}
1
1
1
true
{"x0": 0.7202568481086095, "y0": 0.37092305345389287, "x1": 0.7451405151468428, "y1": 0.24860055643873358}
circle on flat_stove_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_1/frames/t0_rgb.png
false
{ "bytes": [ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 4, ...
spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_1/C1#1
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_1
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
put this over there
L3
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.11087328940629959, "cy": 0.4013538062572479, "r": 0.2878306865692139}
1
1
1
true
{"x0": 0.7202568481086095, "y0": 0.37092305345389287, "x1": 0.7451405151468428, "y1": 0.24860055643873358}
circle on wooden_cabinet_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_1/frames/t0_rgb.png
false
{ "bytes": [ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 4, ...
spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_10/C1#0
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_10
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
pick up this
L2
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.767578125, "cy": 0.265625, "r": 0.10611109852790833}
1
1
1
true
{"x0": 0.7488882538750253, "y0": 0.3599601278117862, "x1": 0.7639332831165537, "y1": 0.2568665760750859}
circle on plate_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_10/frames/t0_rgb.png
false
{ "bytes": [ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 4, ...
spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_10/C1#1
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_10
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
put this over there
L3
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.767578125, "cy": 0.265625, "r": 0.10611109852790833}
1
1
1
true
{"x0": 0.7488882538750253, "y0": 0.3599601278117862, "x1": 0.7639332831165537, "y1": 0.2568665760750859}
circle on plate_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_10/frames/t0_rgb.png
false
{ "bytes": [ 255, 216, 255, 224, 0, 16, 74, 70, 73, 70, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 255, 219, 0, 67, 0, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 2, 2, 2, 2, 2, 4, ...
"spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_(...TRUNCATED)
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_11
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
put this over there
L3
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.11087328940629959, "cy": 0.4013538062572479, "r": 0.2878306865692139}
1
1
1
true
"{\"x0\": 0.7231969087543102, \"y0\": 0.38156177148558856, \"x1\": 0.7630824154246107, \"y1\": 0.272(...TRUNCATED)
"circle on wooden_cabinet_1 but arrow originates at akita_black_bowl_1 — each channel plausible al(...TRUNCATED)
"extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_p(...TRUNCATED)
false
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBw(...TRUNCATED)
"spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_(...TRUNCATED)
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_11
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
put this there
L2
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.55859375, "cy": 0.244140625, "r": 0.0728477954864502}
1
1
1
true
"{\"x0\": 0.7231969087543102, \"y0\": 0.38156177148558856, \"x1\": 0.7630824154246107, \"y1\": 0.272(...TRUNCATED)
circle on cookies_1 but arrow originates at akita_black_bowl_1 — each channel plausible alone
"extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_p(...TRUNCATED)
false
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBw(...TRUNCATED)
"spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_(...TRUNCATED)
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_13
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
pick up this
L2
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.7094253897666931, "cy": 0.5052923560142517, "r": 0.05649170517921448}
1
1
1
true
"{\"x0\": 0.7325369054908009, \"y0\": 0.3762579658698022, \"x1\": 0.7668925716991145, \"y1\": 0.2517(...TRUNCATED)
"circle on glazed_rim_porcelain_ramekin_1 but arrow originates at akita_black_bowl_1 — each channe(...TRUNCATED)
"extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_p(...TRUNCATED)
false
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBw(...TRUNCATED)
"spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate/demo_(...TRUNCATED)
spatial
00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_plate
demo_13
C1
C1
C
Cross-modal consistency
hard
vlm
curated
wrong
false
true
wrong because the visual and language channels are inconsistent
true
true
true
false
true
1
put this there
L2
1
1
1
ref=circle,dst=arrow-head
{"cx": 0.11087328940629959, "cy": 0.4013538062572479, "r": 0.2878306865692139}
1
1
1
true
"{\"x0\": 0.7325369054908009, \"y0\": 0.3762579658698022, \"x1\": 0.7668925716991145, \"y1\": 0.2517(...TRUNCATED)
"circle on wooden_cabinet_1 but arrow originates at akita_black_bowl_1 — each channel plausible al(...TRUNCATED)
"extracted/spatial/00_pick_up_the_black_bowl_between_the_plate_and_the_ramekin_and_place_it_on_the_p(...TRUNCATED)
false
{"bytes":"/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAIBAQEBAQIBAQECAgICAgQDAgICAgUEBAMEBgUGBgYFBgYGBwkIBgcJBw(...TRUNCATED)
End of preview. Expand in Data Studio

DrawVLA — Sketch-Prompt Validation

Circle (which) + arrow (where) + caption (what) visual instructions overlaid on LIBERO observations, each labelled with a binary verdict for training a prompt validator or a self-checking VLA:

  • right — every channel is correct and exactly one reading survives; execute.
  • wrong — a channel is incorrect or the deictic prompt remains under-determined; reject. Formerly ambiguous prompts are retained in this class.

All captions are name-free L2/L3 deixis (for example, "move this there"). Object identity must come from the circle/arrow sketch, not an explicit noun. n_readings = n_referents x n_destinations is retained as diagnostic metadata. Embedded images use JPEG encoding.

Train and validation rows include label_reason, a human-readable explanation derived from the channel checks and declared grounding count. The demo-disjoint test split keeps its ground-truth verdict for scoring but sets reason_available=false and nulls all reason/rationale fields (label_reason, note, mode/category, failure-channel flags, severity, and resolution diagnostics), preventing rationale leakage during final testing.

Generated from the failure taxonomy (failure_modes.md). Split into 7 configs by family; each loads independently:

from datasets import load_dataset
ds = load_dataset("shibuina/drawvla-prompt-validation", "crossmodal", split="train")

Configs (37,988 samples)

config train validation test total right / wrong
crossmodal 1,212 150 158 1,520 0 / 1,520
directional 7,338 904 958 9,200 0 / 9,200
grounding 5,648 695 655 6,998 0 / 6,998
language 2,556 308 336 3,200 0 / 3,200
physical 14 0 4 18 0 / 18
positive 7,194 879 1,025 9,098 9,098 / 0
referential 6,354 766 834 7,954 0 / 7,954

Fields

image, id, mode, category, tier, solution, verdict, right, reason_available, label_reason, circle_ok/arrow_ok/caption_ok/consistency_ok/sufficient_ok, severity, caption, caption_level, n_referents/n_destinations/n_readings, resolved_by, grounding_kind, rendered_n_readings/render_verified, circle_json, arrow_json, name_explicit, note. The sketch geometry (circle_json, arrow_json, image-plane, normalized) lets you re-render at any resolution. This release samples only L2 (relational deixis) and L3 (pure deixis); L0/L1 are excluded because they explicitly name the task or object category. Splits are approximately 80/10/10 by demo; no demo straddles train, validation, or test.

Built with the pipeline in data_gen/ on the OpenVLA modified_libero_rlds demos.

Downloads last month
230